sched: updating /proc/schedstat

sched: Updating the fields of /proc/schedstat

This patch removes the unused statistics from /proc/schedstat. Also updates the request queue structure fields.

Signed-off-by Ciju Rajan K<[email protected]>

include/linux/sched.h +0 -11

index d747f94..1c0ac12 100644
@@ -954,20 +954,9 @@ struct sched_domain {
  	unsigned int alb_failed;
  	unsigned int alb_pushed;
 
-	/* SD_BALANCE_EXEC stats */
-	unsigned int sbe_count;
-	unsigned int sbe_balanced;
-	unsigned int sbe_pushed;
-
-	/* SD_BALANCE_FORK stats */
-	unsigned int sbf_count;
-	unsigned int sbf_balanced;
-	unsigned int sbf_pushed;
-
  	/* try_to_wake_up() stats */
  	unsigned int ttwu_wake_remote;
  	unsigned int ttwu_move_affine;
-	unsigned int ttwu_move_balance;
  #endif
  #ifdef CONFIG_SCHED_DEBUG
  	char *name;

kernel/sched_debug.c +0 -1

index eb6cb8e..99893be 100644
@@ -286,7 +286,6 @@ static void print_cpu(struct seq_file *m, int cpu)
 
  	P(yld_count);
 
-	P(sched_switch);
  	P(sched_count);
  	P(sched_goidle);
  #ifdef CONFIG_SMP

kernel/sched_stats.h +5 -9

index 48ddf43..8869ed9 100644
@@ -4,7 +4,7 @@
   * bump this up when changing the output format or the meaning of an existing
   * format, so that tools can adapt (or abort)
   */
-#define SCHEDSTAT_VERSION 15
+#define SCHEDSTAT_VERSION 16
 
  static int show_schedstat(struct seq_file *seq, void *v)
  {

@@ -26,9 +26,9 @@ static int show_schedstat(struct seq_file *seq, void *v)
 
  		/* runqueue-specific stats */
  		seq_printf(seq,
-		    "cpu%d %u %u %u %u %u %u %llu %llu %lu",
+		    "cpu%d %u %u %u %u %u %llu %llu %lu",
  		    cpu, rq->yld_count,
-		    rq->sched_switch, rq->sched_count, rq->sched_goidle,
+		    rq->sched_count, rq->sched_goidle,
  		    rq->ttwu_count, rq->ttwu_local,
  		    rq->rq_cpu_time,
  		    rq->rq_sched_info.run_delay, rq->rq_sched_info.pcount);

@@ -57,12 +57,9 @@ static int show_schedstat(struct seq_file *seq, void *v)
  				    sd->lb_nobusyg[itype]);
  			}
  			seq_printf(seq,
-				   " %u %u %u %u %u %u %u %u %u %u %u %u\n",
+				   " %u %u %u %u %u\n",
  			    sd->alb_count, sd->alb_failed, sd->alb_pushed,
-			    sd->sbe_count, sd->sbe_balanced, sd->sbe_pushed,
-			    sd->sbf_count, sd->sbf_balanced, sd->sbf_pushed,
-			    sd->ttwu_wake_remote, sd->ttwu_move_affine,
-			    sd->ttwu_move_balance);
+			    sd->ttwu_wake_remote, sd->ttwu_move_affine);
  		}
  		preempt_enable();
  #endif
--

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

This patch is logically correct, succeeded to compile and works fine. But I came to be worried about whether it is good to kill all fields you said after reading old and upstream scheduler code again.

I think we can remove rq->sched_switch and rq->sched_switch without no problem because they are meaningless. The former is for old O(1) scheduler and means the number of runqueue switching among active/expired queue. The latter is for SD_WAKE_BALANCE flag and its logic is already gone.

However sbe_* are for SD_BALANCE_EXEC flag and sbf_* are for SD_BALANCE_FORK flag. Since both logic for them are still alive, the absence of these accounting is regression in my perspective. In addition, these fields would be useful for analyzing load balance behavior.

# although I haven't been able to notice they are always zero ;-(

I prefer not to remove these fields({sbe,sbf}_*) but to add accounting code for these flags again. What do you think?

Quoted 6 lines

I think we can remove rq->sched_switch and rq->sched_switch without no problem because they are meaningless. The former is for old O(1) scheduler and means the number of runqueue switching among active/expired queue. The latter is for SD_WAKE_BALANCE flag and its logic is already gone.

Quoted 6 lines

However sbe_* are for SD_BALANCE_EXEC flag and sbf_* are for SD_BALANCE_FORK flag. Since both logic for them are still alive, the absence of these accounting is regression in my perspective. In addition, these fields would be useful for analyzing load balance behavior.

Hi Satoru,

Quoted 6 lines

Quoted 6 lines

I think we can remove rq->sched_switch and rq->sched_switch without no problem because they are meaningless. The former is for old O(1) scheduler and means the number of runqueue switching among active/expired queue. The latter is for SD_WAKE_BALANCE flag and its logic is already gone.

Quoted 6 lines

Quoted 6 lines

However sbe_* are for SD_BALANCE_EXEC flag and sbf_* are for SD_BALANCE_FORK flag. Since both logic for them are still alive, the absence of these accounting is regression in my perspective. In addition, these fields would be useful for analyzing load balance behavior.

The sbe_*& sbf_* counters were added by the commit 68767a0ae428801649d510d9a65bb71feed44dd1 But it was subsequently removed by the commit 476d139c218e44e045e4bc6d4cc02b010b343939

[ciju@ciju kernel]$ git describe 68767a0ae428801649d510d9a65bb71feed44dd1 --contains v2.6.13-rc1~68^2~148
[ciju@ciju kernel]$ git describe 476d139c218e44e045e4bc6d4cc02b010b343939 --contains v2.6.13-rc1~68^2~140

So.. it was introduced and removed in 2.6.13 time frame

When the counters were removed the sbe_* sbf_* variable declarations were not removed. Hence it caused a little confusion. So I believe these stats were not available and hence can't be considered as regression.

476d139c218e44e045e4bc6d4cc02b010b343939 consolidated the fork and exec balance. Thereafter it became non-trivial to provide separate stats for fork and exec events. So if people think a consolidated balance-on-event is needed, it can be looked into separately. But that shouldn't prevent this documentation cleanup patch from getting in.

-Ciju

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Hi Ciju,

Quoted 16 lines · Ciju Rajan K

Hi Satoru,

Quoted 6 lines

Quoted 6 lines

I think we can remove rq->sched_switch and rq->sched_switch without no problem because they are meaningless. The former is for old O(1) scheduler and means the number of runqueue switching among active/expired queue. The latter is for SD_WAKE_BALANCE flag and its logic is already gone.

Quoted 6 lines

Quoted 6 lines

However sbe_* are for SD_BALANCE_EXEC flag and sbf_* are for SD_BALANCE_FORK flag. Since both logic for them are still alive, the absence of these accounting is regression in my perspective. In addition, these fields would be useful for analyzing load balance behavior.

Quoted 3 lines

The sbe_*& sbf_* counters were added by the commit 68767a0ae428801649d510d9a65bb71feed44dd1 But it was subsequently removed by the commit 476d139c218e44e045e4bc6d4cc02b010b343939

OK, I understood. It's OK if user tools referring /proc/schedstat are released sync with this change.

I confirmed the following:

- This patch removes some unused schedstat fields and related data.
- The kernel applying this patch works fine on my i386 box.

Tested-by Satoru Takeuchi <[email protected]>

Thanks,
Satoru

Quoted 31 lines

[ciju@ciju kernel]$ git describe 68767a0ae428801649d510d9a65bb71feed44dd1 --contains v2.6.13-rc1~68^2~148
[ciju@ciju kernel]$ git describe 476d139c218e44e045e4bc6d4cc02b010b343939 --contains v2.6.13-rc1~68^2~140

So.. it was introduced and removed in 2.6.13 time frame

When the counters were removed the sbe_* sbf_* variable declarations were not removed. Hence it caused a little confusion. So I believe these stats were not available and hence can't be considered as regression.

476d139c218e44e045e4bc6d4cc02b010b343939 consolidated the fork and exec balance. Thereafter it became non-trivial to provide separate stats for fork and exec events. So if people think a consolidated balance-on-event is needed, it can be looked into separately. But that shouldn't prevent this documentation cleanup patch from getting in.

-Ciju

Signature · 4 lines

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

sched: Updating the sched-stat documentation

Some of the unused fields are removed from /proc/schedstat. This is the documentation changes reflecting the same.

Signed-off-by Ciju Rajan K<[email protected]>

---

Documentation/scheduler/sched-stats.txt +56 -95

index 01e6940..28bee75 100644
@@ -26,119 +26,81 @@ Note that any such script will necessarily be version-specific, as the main
  reason to change versions is changes in the output format.  For those wishing
  to write their own scripts, the fields are described here.
 
+The first two fields of /proc/schedstat indicates the version (current
+version is 16) and jiffies values. The following values are from
+cpu&  domain statistics.
+
  CPU statistics
  --------------
-cpu<N>  1 2 3 4 5 6 7 8 9 10 11 12
-
-NOTE: In the sched_yield() statistics, the active queue is considered empty
-    if it has only one process in it, since obviously the process calling
-    sched_yield() is that process.
-
-First four fields are sched_yield() statistics:
-     1) # of times both the active and the expired queue were empty
-     2) # of times just the active queue was empty
-     3) # of times just the expired queue was empty
-     4) # of times sched_yield() was called
-
-Next three are schedule() statistics:
-     5) # of times we switched to the expired queue and reused it
-     6) # of times schedule() was called
-     7) # of times schedule() left the processor idle
+The format is like this:
 
-Next two are try_to_wake_up() statistics:
-     8) # of times try_to_wake_up() was called
-     9) # of times try_to_wake_up() was called to wake up the local cpu
+cpu<N>  1 2 3 4 5 6 7 8
 
-Next three are statistics describing scheduling latency:
-    10) sum of all time spent running by tasks on this processor (in jiffies)
-    11) sum of all time spent waiting to run by tasks on this processor (in
-        jiffies)
-    12) # of timeslices run on this cpu
+     1) # of times sched_yield() was called on this CPU
+     2) # of times scheduler runs on this CPU
+     3) # of times scheduler picks idle task as next task on this CPU
+     4) # of times try_to_wake_up() is run on this CPU
+        (Number of times task wakeup is attempted from this CPU)
+     5) # of times try_to_wake_up() wakes up a task on the same CPU
+        (local wakeup)
+     6) Time(ns) for which tasks have run on this CPU
+     7) Time(ns) for which tasks on this CPU's runqueue have waited
+        before getting to run on the CPU
+     8) # of tasks that have run on this CPU
 
 
  Domain statistics
  -----------------
-One of these is produced per domain for each cpu described. (Note that if
-CONFIG_SMP is not defined, *no* domains are utilized and these lines
-will not appear in the output.)
+One of these is produced per domain for each cpu described.
+(Note that if CONFIG_SMP is not defined, *no* domains are utilized
+ and these lines will not appear in the output.)
 
-domain<N>  <cpumask>  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
+domain<N>  <cpumask>  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
 
  The first field is a bit mask indicating what cpus this domain operates over.
 
-The next 24 are a variety of load_balance() statistics in grouped into types
-of idleness (idle, busy, and newly idle):
-
-     1) # of times in this domain load_balance() was called when the
-        cpu was idle
-     2) # of times in this domain load_balance() checked but found
-        the load did not require balancing when the cpu was idle
-     3) # of times in this domain load_balance() tried to move one or
-        more tasks and failed, when the cpu was idle
-     4) sum of imbalances discovered (if any) with each call to
-        load_balance() in this domain when the cpu was idle
-     5) # of times in this domain pull_task() was called when the cpu
-        was idle
-     6) # of times in this domain pull_task() was called even though
-        the target task was cache-hot when idle
-     7) # of times in this domain load_balance() was called but did
-        not find a busier queue while the cpu was idle
-     8) # of times in this domain a busier queue was found while the
-        cpu was idle but no busier group was found
-
-     9) # of times in this domain load_balance() was called when the
-        cpu was busy
-    10) # of times in this domain load_balance() checked but found the
-        load did not require balancing when busy
-    11) # of times in this domain load_balance() tried to move one or
-        more tasks and failed, when the cpu was busy
-    12) sum of imbalances discovered (if any) with each call to
-        load_balance() in this domain when the cpu was busy
-    13) # of times in this domain pull_task() was called when busy
-    14) # of times in this domain pull_task() was called even though the
-        target task was cache-hot when busy
-    15) # of times in this domain load_balance() was called but did not
-        find a busier queue while the cpu was busy
-    16) # of times in this domain a busier queue was found while the cpu
-        was busy but no busier group was found
-
-    17) # of times in this domain load_balance() was called when the
-        cpu was just becoming idle
-    18) # of times in this domain load_balance() checked but found the
-        load did not require balancing when the cpu was just becoming idle
-    19) # of times in this domain load_balance() tried to move one or more
-        tasks and failed, when the cpu was just becoming idle
-    20) sum of imbalances discovered (if any) with each call to
-        load_balance() in this domain when the cpu was just becoming idle
-    21) # of times in this domain pull_task() was called when newly idle
-    22) # of times in this domain pull_task() was called even though the
-        target task was cache-hot when just becoming idle
-    23) # of times in this domain load_balance() was called but did not
-        find a busier queue while the cpu was just becoming idle
-    24) # of times in this domain a busier queue was found while the cpu
-        was just becoming idle but no busier group was found
-
+The next 24 are a variety of load_balance() statistics grouped into
+types of idleness (idle, busy, and newly idle). The three idle
+states are:
+
+CPU_IDLE:          This state is entered after CPU_NEWLY_IDLE
+                   state fails to find a new task for this CPU
+CPU_NOT_IDLE:      Load balancer is being run on a CPU when it is
+                   not in IDLE state (busy times)
+CPU_NEWLY_IDLE:    Load balancer is being run on a CPU which is
+                   about to enter IDLE state
+
+There are eight stats available for each of the above three states:
+     - # of times in this domain load_balance() was called
+     - # of times in this domain load_balance() checked but found
+        the load did not require balancing
+     - # of times in this domain load_balance() tried to move one or
+        more tasks and failed
+     - sum of imbalances discovered (if any) with each call to
+        load_balance() in this domain
+     - # of times in this domain pull_task() was called
+     - # of times in this domain pull_task() was called even though
+        the target task was cache-hot
+     - # of times in this domain load_balance() was called but did
+        not find a busier queue
+     - # of times in this domain a busier queue was found but no
+        busier group was found
+
+   The first 1-8) fields are the stats when cpu was idle (CPU_IDLE),
+   the next 9-15) fields are the stats when cpu was busy (CPU_NOT_IDLE),
+   and the next 16-24) fields are the stats when cpu  was just
+   becoming idle (CPU_NEWLY_IDLE)
+
     Next three are active_load_balance() statistics:
      25) # of times active_load_balance() was called
      26) # of times active_load_balance() tried to move a task and failed
      27) # of times active_load_balance() successfully moved a task
 
-   Next three are sched_balance_exec() statistics:
-    28) sbe_cnt is not used
-    29) sbe_balanced is not used
-    30) sbe_pushed is not used
-
-   Next three are sched_balance_fork() statistics:
-    31) sbf_cnt is not used
-    32) sbf_balanced is not used
-    33) sbf_pushed is not used
-
-   Next three are try_to_wake_up() statistics:
-    34) # of times in this domain try_to_wake_up() awoke a task that
+   Next two are try_to_wake_up() statistics:
+    28) # of times in this domain try_to_wake_up() awoke a task that
          last ran on a different cpu in this domain
-    35) # of times in this domain try_to_wake_up() moved a task to the
+    29) # of times in this domain try_to_wake_up() moved a task to the
          waking cpu because it was cache-cold on its own cpu anyway
-    36) # of times in this domain try_to_wake_up() started passive balancing
 
  /proc/<pid>/schedstat
  ----------------
--

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Hi All,

Here is the v3.0 of the patch set, which updates /proc/schedstat statistics. Please review the patches and consider for inclusion.